Challenge: Ultra-Low-Power Energy-Harvesting Active
Networked Tags (EnHANTs)
Maria Gorlatova†, Peter Kinget†, Ioannis Kymissis†,
Dan Rubenstein∗, Xiaodong Wang†, Gil Zussman†
†Electrical
Engineering,∗Computer Science
Columbia University, New York, NY, 10027
[email protected], [kinget, johnkym]@ee.columbia.edu
[email protected], [wangx, gil]@ee.columbia.edu
This paper presents the design challenges posed by a new class
of ultra-low-power devices referred to as Energy-Harvesting Active Networked Tags (EnHANTs). EnHANTs are small, flexible,
and self-reliant (in terms of energy) devices that can be attached to
objects that are traditionally not networked (e.g., books, clothing,
and produce), thereby providing the infrastructure for various novel
tracking applications. Examples of these applications include locating misplaced items, continuous monitoring of objects (items in
a store, boxes in transit), and determining locations of disaster survivors. Recent advances in ultra-low-power wireless communications, ultra-wideband (UWB) circuit design, and organic electronic
harvesting techniques will enable the realization of EnHANTs in
the near future. In order for EnHANTs to rely on harvested energy,
they have to spend significantly less energy than Bluetooth, Zigbee, and IEEE 802.15.4a devices. Moreover, the harvesting components and the ultra-low-power physical layer have special characteristics whose implications on the higher layers have yet to be
studied (e.g., when using ultra-low-power circuits, the energy required to receive a bit is an order of magnitude higher than the energy required to transmit a bit). These special characteristics pose
several new cross-layer research problems. In this paper, we describe the design challenges at the layers above the physical layer,
point out relevant research directions, and outline possible starting
points for solutions.
Categories and Subject Descriptors
C.2.1 [Computer Communication Networks]: Network Architecture and Design - Wireless communication
General Terms
Algorithms, Design, Performance
Keywords
Ultra-low power communications, energy efficient networking, energy harvesting, energy scavenging, ultra-wideband (UWB)
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
MobiCom’09, September 20–25, 2009, Beijing, China.
Copyright 2009 ACM 978-1-60558-702-8/09/09 ...$10.00.
Sensor Networks
EnHANTs
RFIDs
Throughput
Complexity
Size
Energy consumption
ABSTRACT
Figure 1: EnHANTs in comparison to Sensor Networks and
RFIDs.
1. INTRODUCTION
This paper focuses on the networking challenges posed by a
new class of ultra-low-power devices that we refer to as EnergyHarvesting Active Networked Tags (EnHANTs). EnHANTs are
small, flexible, and self-reliant (in terms of energy) devices that can
be attached to arbitrary objects that are traditionally not networked:
books, clothing, produce, etc. EnHANTs will enable novel object
tracking applications such as recovery of lost items and continuous monitoring of objects’ proximity to each other. The realization of EnHANTs is based on recent advances in the areas of solar
and piezoelectric energy harvesting [24] as well as ultra-low-power
wireless communications [26, 42]. In particular, recent novel circuit designs that employ ultra-wideband (UWB) communications
provide new levels of ultra-low-power operation (at the orders of
nJ/bit) at short ranges. Moreover, solar energy harvesting based on
organic semiconductors allows having flexible solar panels [19,25],
thereby allowing a pervasive use of tags.
The wireless industry is already taking the first steps towards the
design of energy harvesting ultra-low-power tags [2,3]. Hence, following the transition from barcodes to RFIDs, we envision a future
transition from RFIDs to EnHANTs that:
• Network – Actively communicate with one another and with
EnHANT-friendly devices in order to forward information
over a multihop network.
• Operate at ultra-low-power – Spend a few nano-Joules or
less on every transmitted bit.
• Harvest energy – Collect and store energy from sources
such as light, motion, and temperature gradients.
• Are energy adaptive – Alter communications and networking to satisfy energy and harvesting constraints.
• Exchange small messages – Exchange limited information
(basically IDs) using low data rates, possibly in several transmission bursts.
• Transmit to short ranges – Communicate only when in close
proximity (1 to 10 meters) to one another.
• Are thin, flexible, and small (a few square cm at most).
As shown in Figure 1, in terms of complexity, throughput, size,
and energy requirements, EnHANTs fit between RFIDs and sensor networks. Similarly to RFIDs, the tags can be affixed to commonplace objects. However, EnHANTs will have a power source,
will be able to communicate in distributed multihop fashion, and
will not have to rely on high-power readers. Compared to sensor nodes, EnHANTs will operate at significantly lower data rates,
and will consume less energy. Moreover, unlike sensor nodes, EnHANTs will transmit mostly ID information (either in order to announce themselves or to query for specific EnHANTs). Despite
these differences, some of the results obtained for sensor networks
(see [14, 16]) should apply to EnHANTs.
EnHANTs will be enablers for the Internet of Things and as such
will support a variety of tracking and monitoring applications beyond what RFID permits. While RFIDs make it possible to identify
an object, EnHANTs will make it possible to search for an object,
and to continuously track objects’ whereabouts and their proximity to each other. RFIDs are typically activated only when placed
near a reader, and only report on themselves. EnHANTs, on the
other hand, can operate continuously, achieve pervasive coverage
due to their networking capabilities, and can report on themselves
and other EnHANTs around them. These EnHANTs capabilities
enable many exciting applications, such as, for example, continuous peer monitoring of merchandize in transit, where EnHANTs
would be able to identify if a particular box has been taken out at
any point during the journey.
One application that we plan to demonstrate in the near future is
a misplaced library book locator. The initial prototype will enable
library books to identify those among themselves that are significantly misplaced (e.g., in an incorrect section), and report the misplacement. To accomplish this task, each book is assigned a unique
ID using an assignment scheme closely related to the Dewey Decimal Classification. Each book has a solar powered tag whose power
output is sufficient to transmit and receive information within a
radius of one meter or less, and to perform some basic processing. Nearby books wirelessly exchange IDs, and IDs of books that
appear out of place are further forwarded through the network of
books, eventually propagating to sink nodes. A long term objective
could be to place books in an arbitrary order, and both determine
the book order, and propagate that information to a central server
using the multihop wireless network. This type of system can significantly simplify the organization of physical objects.
The same building blocks used in the library application can enable several other applications. In particular, a large variety of items
can be tracked and a range of possible desirable or undesirable configurations of objects can be queried for, and can trigger reports.
Examples include finding items with particular characteristics in a
store, locating misplaced items (e.g., keys or eyeglasses), and locating survivors of disasters such as structural collapse [45].
To enable these applications, various protocols have to be designed. Although networking protocols for energy harvesting nodes
recently started gaining attention [6,8,9,11–13,18,22,23,27,35,37,
38] (see Section 2), to the best of our knowledge, the cross layer interactions between circuit design, energy harvesting, communications, and networking have not been studied in depth. Current RF
transceiver designs, communication and networking protocols, and
energy harvesting and management techniques have been developed in isolation and are inadequate for the envisioned EnHANTs
applications. Hence, based on our experience with hardware de-
sign, communications, and networking, we outline the cross-layer
design challenges that are posed by this new technology. We note
that although there are many hardware-specific design challenges,
they are out of scope for this paper.
EnHANTs are likely to be implemented by combining flexible
electronics technologies (a.k.a. organic electronics) with CMOS
chips supporting Impulse-Radio UWB. Flexible technologies can
realize energy harvesters (solar cells, piezoelectric, thermal, etc.),
passive RF components, and batteries. By embedding CMOS chips
in EnHANTs, very low power computation, memory, and communication functions can be added. This technology platform allows
for thin, flexible, and very low cost EnHANT fabrication. By using
Impulse-Radio UWB, data is encoded by very short pulses (at the
order of nano-seconds) and the energy consumption is very low.
To better understand the higher layer design challenges, we first
discuss energy harvesting and energy storage techniques. Energy
storage is required in order to use the harvested energy in periods
in which harvesting is not possible. We later describe models that
take these techniques and their characteristics into account when
determining the UWB pulse patterns, duty cycles, and overall energy consumption. Some of the interesting characteristics include
the differences in performance between a capacitor and a battery,
and between indoor and outdoor solar energy harvesting.
Next, we discuss ultra low power UWB communications and
present a number of design challenges. These include a paradigm
shift resulting from the fact that when using ultra-low-power communications it is energetically cheaper to transmit data than to receive data. Moreover, ultra-low-power tags will operate with inaccurate clocks, thereby requiring the redesign of low-power methods
based on coordination of wakeup periods. Finally, EnHANTs may
have additional circuitry (e.g., accurate clocks) that can be powered up in some harvesting states and can be used to assist other
EnHANTs. Determining how and when to use this circuitry is another challenge.
We then focus on the design of EnHANTs’ communications and
networking protocols. These protocols have to determine the state
of the tags (sleeping, communicating, etc.), the operation within a
state (transmit, receive, rate control, etc.), and the coordination with
peer tags. All these have to be based on the harvesting states of the
tags and their capabilities. Clearly, energy harvesting shifts the nature of energy-aware protocols from prolonging the finite lifespan
of a device to enabling perpetual life, and from minimizing energy
expenditure to optimizing it. The challenges that we describe include not only determining the state and the operation model within
the state but also the use of a “harvesting channel” as a mean for
nodes synchronization.
Finally, we draw parallels between energy-harvesting EnHANTs
and large-scale manufacturing systems. We show that the harvested
energy can be treated similarly to inventory in production and storage systems. We show how inventory control models extensively
studied in the inventory management field [32, 33] apply to EnHANTs. We then discuss the various challenges resulting from
the fact that EnHANTs compose a distributed network with many
stochastic components, and outline a number of open problems.
We are currently building a testbed of EnHANTs in which we
will evaluate various approaches using energy harvesting and ultralow-power hardware. We conclude by briefly describing the implementation phases and the hardware components.
This paper is organized as follows. In Section 2 we describe
related work. In Sections 3 and 4 we discuss energy harvesting, energy storage, ultra-low-power communications, and the challenges
they pose for designing higher layer protocols. Section 5 describes
the challenges posed by EnHANTs communications and network-
ing, and Section 6 discusses the application of inventory control
models to EnHANTs. Finally, in Section 7 we present our plan for
designing an EnHANTs testbed.
2.
RELATED WORK
While the idea of pervasive networks of objects has been proposed before (e.g., in the Smart Dust project [41]), the harvesting and communications technologies have recently reached a point
where networked energetically self-reliant tags are becoming practical. As mentioned above, EnHANTs fall between RFIDs [40]
and sensor networks. To our knowledge, most of the networking
research in the area of RFID focuses on the scheduling of query
responses by passive tags (e.g., [15]). Extensively studied sensor
networks are designed to deal with energy, bandwidth, and other
resource constraints (see [14, 16] and references therein). Yet, the
underlying communication mechanisms draw power at rates that
are too high in environments with weak energy sources (indoor
lighting, strain, vibration).
EnHANTs will employ Impulse Radio UWB communications
[43], whose corresponding MAC protocols (e.g., [20,31]) have been
proposed for communications and accurate ranging in sensor networks. The recent IEEE 802.15.4a standard [4] is based on Impulse Radio UWB and inherits many of the IEEE 802.15.4 (ZigBee) functionalities. Limited hardware data is currently available,
but we can assume that the overhead to provide ranging, high data
rates (up to 26Mb/s), and backward compatibility will lead to energy consumption largely exceeding the energy we envision available in EnHANTs.
Energy efficiency in wireless networks has long been a subject of
research (see reviews [5, 10, 21]). In comparison, only a few works
have considered energy-harvesting. One of the major directions
in exploiting energy harvesting is adaptive duty cycling in sensor
networks. In particular, [8, 11, 12] describe adaptation of the duty
cycle to the characteristics of a periodic energy source and [37] develops a “battery-centric” (not requiring an energy source model)
duty-cycle adaptation algorithm. Taking energy harvesting into account when making routing decisions is studied in [18, 38, 39]. Recently, [6] jointly calculates link flows and data collection rates for
energy-harvesting sensor networks.
Dynamic activation of energy-harvesting sensors has been studied in [9, 13, 22, 23]. A system where mobile nodes deliver harvested energy to different areas of the network is presented in [27].
A system where an energy-restricted node can “outsource” packet
retransmissions is described in [34]. A power subsystem for HydroWatch is described in [35], which also provides an overview of deployments that rely on energy harvesting. To the best of our knowledge, current deployments use much more energy than EnHANTs
will have available.
Finally, there is an increasing industry interest in bringing together low-power communications and energy harvesting (e.g., [2,
28]). Particularly, Texas Instruments has recently put on the market a solar energy harvesting development kit [3] geared towards
sensor developments. Although it is a first step towards EnHANTs,
this kit is made of rigid materials and is more than 10x the size of
the EnHANTs envisioned in this work.
3.
ENERGY HARVESTING
In order to outline the higher layer design challenges, we briefly
describe two major harvesting methods and possible energy storage
methods. Then, we discuss the effects of the different methods on
the design and operation of higher layer protocols. In Sections 5
Figure 2: An organic semiconductor-based small molecule solar cell series array developed in the Columbia Laboratory for
Unconventional Electronics (CLUE).
and 6 we discuss different approaches to the design of such higher
layer protocols.
3.1 Energy Sources
Many environmental sources of energy are potentially available
for harvesting by small devices. These include temperature differences, electromagnetic energy, airflow, and vibrations [24, 30].
Below, we focus on the most promising harvesting technologies
for EnHANTs: solar energy and piezoelectric (motion) harvesting.
Other harvesting technologies pose similar design challenges.
Solar energy (light) is one of the most useful energy sources,
with typical irradiance (total energy projected and available for collection) ranging from 100mW/cm2 in direct sunlight to 0.1mW/cm2
in brightly lit residential indoor environments (notice the significant
difference) [24, 29]. Office, retail, and laboratory environments are
typically brighter than residential settings, but get much less light
energy than outdoor environments. The efficiency of a solar energy harvesting device is defined as the percentage of the available energy that is actually harvested. Conventional single crystal
and polycrystalline solar cells, such as those that are commonly
used in calculators, have efficiency of around 10%-20% in direct
sunlight [7]. However, their efficiency declines with a decline in
energy availability (they are less efficient with dimmer sources),
which is important to note due to considerable irradiance difference
between direct sunlight and indoor illumination. Conventional solar panels are also inflexible (rigid), which makes it difficult to attach them to non-rigid items such as clothing and paperback books.
An emerging and less explored option is solar energy harvesting based on organic semiconductors [19, 25] (an array of organic
solar cells that we recently designed is shown in Figure 2). With
this technology, solar cells can be made flexible. Moreover, organic
semiconductor-based panels operate with constant efficiencies over
different brightness levels. However, their efficiency is typically
1%-1.5% [7], which is much lower than the efficiency of conventional inorganic solar panels.
To put the numbers in perspective, consider a system with a
10cm2 organic semiconductor cell. Outdoors, the system will harvest 10cm2 ·100mW/cm2 ·0.01=10mW. Under the assumption that
reception of a single bit requires 1nJ1 , the achievable data rate will
be (10 · 10−3 )/(1 · 10−9 ) = 10Mb/s. The achievable data rate with
indoor lighting will be (10−5 )/(1 · 10−9 ) = 10Kb/s.
Another potential source of energy is piezoelectric (motion) energy. It can be generated by straining a material (e.g., squeezing or bending flexible items). An example is energy harvesting
through footfall, where a harvesting device is placed in a shoe and
piezoelectric energy is generated and captured with each step [17].
1 Recall that reception is more expensive than transmission (for
more details see Section 4).
E
e
r
C
Figure 3: An abstraction of an energy harvesting system for the
upper layers: the energy storage capacity C, the current energy
level E, the energy charge rate r, and the energy consumption
rate e.
Unlike solar harvesting, piezoelectric harvesting may be somewhat
controlled by the user.
Piezoelectric harvesting is characterized by the energy captured
per actuation at a particular strain (usually 1%-3%). In [17] it was
shown how to harvest 4µJ/cm2 per deflection with a strain of approximately 1.5% from straining polyvinylidene fluoride (PVDF),
a highly compliant piezoelectric polymer. Assume that a 10cm2
of material is employed in an environment where it is strained 60
times per second. This would provide 10 · 4 · 10−6 · 60 = 2.4mW. If,
similar to above, we assume that transmission of a bit costs 1nJ, the
bit rate that can be supported is (2.4 · 10−3 )/(1 · 10−9 ) = 2.4Mb/s.
3.2 Energy Storage
Without the ability to store energy, a device can operate only
when directly powered by environmental energy. For a tag, energy
storage components need to be compact and efficient, and need to
have very low self-discharge rates. From a higher-layer point of
view, it is also important to have storage elements that are straightforward to measure and control.
Rechargeable batteries are an excellent option for energy storage, and numerous battery options are available. Thin film batteries
are particularly attractive for EnHANTs since they are environmentally friendly and can be made flexible. However, a battery needs to
be supplied with a voltage exceeding the internal chemical potential
(typically 1.5-3.7V) in order to start storing provided energy. This
implies that charge generated at a low voltage, such as that possibly produced during low harvester excitation (for example, when
a solar cell is located in a very dimly lit place), cannot be stored
without voltage upconversion.
Capacitors can also be used for energy storage. Capacitors can
receive any charge which exceeds their stored voltage and be cycled many more times than batteries. The disadvantage of using
capacitors, however, is that as a capacitor gets more charged, it becomes more difficult to add charge, and large electrolytic capacitors
self-discharge over hours or days. The energy density (how much
energy can be stored per unit of volume) of capacitors is also much
lower. A typical battery can store about 1000J/cm3 , whereas high
performance ceramic capacitors can store 1-10J/cm3 [44].
Different EnHANTs applications will require different types of
energy storage. For example, a tag which frequently experiences
shallow charging and discharging events will need a capacitor for
an acceptable lifetime, whereas an EnHANT that needs to operate for a long period of time without recharging and to store large
amounts of energy (say, to charge all day and discharge all night)
will need the energy density that a battery offers.
3.3 Higher Layer View of Harvesting
The complexity of the harvesting system needs to be captured in
a relatively simple way for the higher layers. A possible abstraction
is shown in Figure 3. The system is characterized by the maximum
energy storage capacity C (Joules), the currently available energy
level E (Joules), the energy charge rate r (Watts), and the energy
consumption rate e (Watts). Note that the energy charge rate r depends both on the harvesting rate and the properties of the energy
storage. For example, when a battery is used, r is positive only
when the voltage at the energy harvesting component exceeds the
internal chemical potential of the battery. When a capacitor is used,
the relationship of r and energy harvesting rate varies with E. The
energy consumption rate e is controlled by higher-layer algorithms
and with low duty cycle is mostly affected by communication and
networking protocols. The effect of these protocols on e will be
discussed in Sections 5 and 6.
The available energy E and energy charge rate r can be measured directly from the energy harvesting and power conditioning
components. The maximum storage capacity C should be known
nominally from the tag design but more precise characterization is
possible. For example, if a battery is used, C can be projected from
the battery’s age, the storage temperature history, and the charge
level over time. As the battery continues to age and cycle through
charge/discharge events, the capacity will predictably decrease.
The r values of different EnHANTs operating in the same environment will be significantly different. Our experiments with commercial hardware show a lot of variability in the harvesting rates of
identical harvesting devices under identical light conditions (i.e.,
identical solar cells in the same location harvest energy at rates that
can differ by about 30%). In addition, the r values will differ for
closely located solar cells due to differences in how the cells are
located with respect to light sources and obstacles [29].
4. LOW POWER COMMUNICATIONS
Ultra-wide band (UWB) impulse radio (IR) is a compelling technology for short range ultra-low-power wireless communications
[4, 43]. It uses very short pulses (on the order of nano-seconds)
that are transmitted at regular time intervals with the data encoded
in the pulse amplitude, phase, frequency, or position. At low data
rates, the short duration of the pulses allows most circuitry in the
transmitter or receiver to be shut down between pulses, resulting in
significant power savings compared to narrow-band systems.
Practical CMOS IR circuits with energy consumption on the order of a nJ per bit have been recently demonstrated. For example, in [42] a UWB receiver and transmitter require 2.5nJ/bit and
43pJ/bit, respectively at a pulse rate of 17Mpulses/s.2 Recent publications [26, 36] as well as our ongoing research demonstrate that
UWB IR transceivers in the 3-5GHz band with data rates in the
100Kbit/s to 1Mbit/s range and a transmitter energy of less than
50pJ/bit and receiver energy of less than 500pJ/bit are within reach.
In this section, we outline our envisioned design of UWB transceivers for EnHANTs and the resulting higher layer challenges.
These customized circuits will support ultra low energy consumption and will be integrated with energy harvesting devices while
supporting networking capabilities.
4.1 Energy Costs - a Paradigm Shift
The first networking challenge emerging from the design of the
ultra-low-power transceivers is that the energy to receive a bit is
much higher than the energy to transmit a bit. This is significantly
different from traditional WLANs, where the energy to transmit
is higher than the energy to receive, and 802.15.4, where they are
on the same order. This requires novel networking algorithms for
EnHANTs, since many legacy algorithms are developed under the
assumption of transmission being more expensive than reception.
2 Energy consumption is measured at a particular pulse rate, since
per-bit parameters differ at different bit rates. Note that at lower
bit rates, the energy per bit can increase due to the impact of fixed,
pulse independent, circuitry.
In conventional systems in which narrow-band modulated sinusoids are transmitted, the transmitter has to be active for the entire
duration of the signal transmission. As mentioned above, in UWB
very short pulses convey information, so the transmitter and receiver can wake up for very short time intervals to generate and
receive pulses, and can sleep between subsequent pulses. The receiver’s energy is mostly spent on running low-noise amplification
and data-detection circuits that are consuming energy whenever the
device listens to the medium. Hence, there is no difference (in
terms of energy) between receiving information and listening to
the medium.
The new energy tradeoffs call for the design of new algorithmic
approaches. For example, in order to take the burden off a receiver,
a transmitter would have to enable a receiver to listen to the medium
for short time intervals (e.g., repeat its pulses many times in such a
way that a receiver listening to a short window of pulses would get
all the information transmitted). In Section 5 we discuss in more
detail the effect of this phenomenon on the design of higher layer
protocols.
4.2 Inaccurate Clocks
Accurate on-chip references or clocks cannot be powered down
and consume a lot of energy; they have to be avoided to achieve
ultra-low-power operation. One viable solution is to use energetically cheap clocks (e.g., clocks available from ultra-low-power ring
oscillators). However, the frequency of such clocks will vary significantly from tag to tag and its stability over time is also poor.
A UWB receiver has to wake up at certain times in order to receive pulses. Determining these times with inaccurate clocks imposes major challenges and hence while inaccurate clocks save energy, they increase the energy spent on reception. Moreover, traditional low-power sleep-wake protocols (see [14, 16]) heavily rely
on the use of accurate time slots. Hence, eliminating the availability
of accurate clocks in a tag requires redesigning protocols that were
originally designed specifically for energy efficient networking.
4.3 A High Power Mode
In some cases it may be beneficial to spend more energy than
what is typically spent by a tag (e.g., when the battery is fully
charged E = C and the tag is harvesting energy). In such cases
a tag can operate in a high-power mode. EnHANT circuitry can
be designed to contain optional power-hungry hardware modules
that would allow, for example, for a more accurate/faster clock to
be turned on. Other components that could be considered are more
sensitive receiver stages, more selective filters, and more elaborate
pulse detection methods. The inclusion of such components will
increase the probability of successful communications but all performance enhancements will require additional power.
The challenge is to realize the benefits of these high power modes
at the higher layers. An example of a clear benefit is a tag with an
accurate clock that helps other tags to synchronize. Another example is running the receiver or transmitter more often to help other
nodes to detect each other and to establish communication. Given
a set of high power physical layer capabilities, numerous questions
arise: what is the right harvesting status to use/stop using each of
these capabilities? How will using these capabilities in one tag affect other tags? Is it enough to have these capabilities in a subset of
the tags and what is the right size of the subset?
5.
COMMUNICATIONS & NETWORKING
We now outline EnHANTs-related communications and networking challenges. Recall that given the power constraints, EnHANTs
will be communicating within ranges of 1 to 10 meters. We iden-
tify three states of pairwise EnHANT communications, and note
that the rate of energy consumption e can be adjusted within each
state, and also by moving between states. We also outline the challenges related to routing, information dissemination, and network
security. Some of the concepts discussed in this section are well
known in sensor networking. Hence, we highlight the new challenges, such as presence of transmit-only devices and making use
of the always-open harvesting channel.
5.1 Pairwise EnHANT Communications
In pairwise EnHANT communications, three states can be identified: independent, paired, and communicating. To control its energy spending, a tag can move between states with respect to each
of its neighbors. In each particular state, a tag can consume different amounts of energy e depending on its own energy parameters
(C, E, r), and, when relevant, on the energy parameters of other EnHANTs involved in communications.
In the independent state a tag does not maintain contact with the
other tag. In this state the tag needs to decide how much energy it
wants to spend on listening to the medium and transmitting pulses
(to enable others to find it). The amount of energy consumed can
be controlled by changing the spacing between transmitted pulses
and listening periods, as well as by changing the overall duty cycle.
If a tag is very low on energy, it could transmit pulses but not
listen to the medium. This “transmit-only” mode is feasible and
logical for EnHANTs, since, as described in Section 4, it is energetically cheaper for a tag to transmit than to listen. Accommodating the presence of such transmit-only devices is an interesting
networking challenge.
To start communicating, EnHANTs need to synchronize with
each other. Two EnHANTs will be able to start pairing when a
pulse burst sent by one tag overlaps with a listening interval of the
other tag, or when one tag overhears another tag’s communications
with third parties. Since EnHANTs’ activity intervals are functions
of their energy levels, the time-to-pair is also a function of the tags’
energy levels.
Once paired, EnHANTs need to remain synchronized, by periodically exchanging short bitstreams. The paired state is similar to low power modes of IEEE 802.11 and Bluetooth. However,
it should be noted that the “keep-alive” messages EnHANTs exchange are short pulse bursts, rather than beacons that include tens
of bytes. The minimum frequency of burst exchanges is limited by
the devices’ clock drifts. If high-power nodes, discussed in Section
4, are present, it would help with both improving the time-to-pair
and keeping devices synchronized.
Communicating EnHANTs need to coordinate their transmissions in order to ensure that they do not run out of energy. To
make joint decisions on communication rates, the EnHANTs need
to exchange information about their energy states. EnHANTs are
so energy-constrained that exchanges of their energy parameters
(C, E, r, e) may be too costly. It is a challenge to determine how
much information EnHANTs should exchange (e.g., exchange their
current C, E, r, e values or current C, E, r, e values along with a set
of predictions of future values) and how frequently the information
exchange should be conducted.
In the communicating state, a tag’s energy consumption e is
closely related to its data rate: lower data rates allow less transmission and listening. EnHANTs need to communicate at ultralow data rates, which necessitates making provisions for EnHANTs
taking pauses between transmissions of bits. This delay tolerance
on the bit level is a challenge for networking protocols that often
consider a packet as an atomic unit.
5.3 Higher-layer Challenges
EnHANTs’ capabilities also influence the design of higher-layer
protocols, such as protocols for routing and information dissemination. A variety of energy-efficient routing schemes proposed for
ad-hoc and sensor networks can serve as starting points for EnHANTs routing. Both information pulling (extracting data through
a query) and information pushing (EnHANTs proactively exchanging information to assist in a pull) should be used. Deciding on
the right levels of push and pull is an interesting problem that has
strong relation to work in caching and peer-to-peer networks.
Security and privacy are very important issues for the proposed
EnHANTs applications. Lightweight techniques that have been designed for sensor networks and for RFIDs can serve as starting
points in EnHANTs security research. Moreover, congestion control and interference resolution techniques for EnHANTs could be
a future research direction. Currently, short communication ranges
and low transmission rates ensure that congestion and interference
are not primary concerns.
6.
ENHANTS AS AN INVENTORY SYSTEM
In section 5 we outlined Enhants communications and networking challenges, and noted that EnHANTs’ joint decisions on energy
management are a large-scale optimization problem. In this section
we introduce inventory control theory as a tool that can potentially
be used to approach this problem.
Much like the harvest of a farmer, the “harvest” of a tag needs to
be carefully managed. In spending their harvest, the farmer and the
tag both make sure they neither waste harvest due to storage space
limitations, nor run out of harvest when it is needed. This type of
problem is not well studied in wireless networking, However, it has
been examined in depth in the mature field of inventory management [32, 33]. In this section we show how concepts developed in
the inventory management field apply to EnHANTs.
An energy-harvesting tag can be viewed as a manufacturing system composed of a factory and a warehouse. Consider the abstraction of the energy harvesting system shown in Figure 3, which
identifies system parameters C, r, e, and E. A real-world factorywarehouse has a finite capacity, a rate at which products are manufactured (supply rate), a rate at which products are purchased (demand rate), and a level of inventory – all of which directly cor-
Energy source
T
Tp
E(t)
d
Battery level
e(t)
A benefit of the harvesting system is that EnHANTs in close
proximity will be subject to common stimuli through their energy
harvesting channels. Examples include lights turning on/off or running with modulated intensity (e.g., the 60Hz variation in fluorescent lighting), or vibrations felt by more than one tag. For instance,
when a light is turned on, a tag can assume that the energy parameters of all its neighbors change, and behave accordingly. Information about the relative similarities or differences between EnHANTs stimuli can provide information about proximity and can
be used for synchronization via a channel which is effectively always open.
In communication with each of its neighbors, a tag decides on
both a state of communication and, in the chosen state, rate of energy consumption e. When many devices are involved in communication, the decisions are far from trivial. EnHANTs’ joint energy
decisions on states and rates are a large-scale optimization problem, and a suitable solution for the problem needs to be calculated
by low-power EnHANTs without extensive exchange of control information. Developing the algorithms that will make this possible
and will take into account the realistic considerations discussed in
Sections 3 and 4 is one of the major challenges for EnHANTs.
r(t)
5.2 Communications of Multiple EnHANTs
Energy spending
0
ec
2000
4000
6000
8000
10000
Figure 4: Application of the Economic Production Quantity
(EPQ) model to the EnHANTs domain with a periodic on/off
energy source: (i) the energy charge rate r(t), (ii) the energy
level E(t), and (iii) the energy consumption rate e(t).
respond to the listed harvesting system parameters. A warehouse
inventory management system strives to ensure that the demand is
met and that there are no shortages. Similarly, the tag’s energy
management system should ensure that it utilizes its resources to
communicate efficiently and does not run out of energy. Below,
we explore the similarities between the EnHANTs domain and the
inventory management domain. We give two examples of direct
applications of inventory management models to the EnHANTs domain, and demonstrate that due to the distributed operation and the
dependencies between EnHANTs parameters (see Section 5), a network composed of EnHANTs is more complicated than a manufacturing system. This calls for the extension of the well-established
inventory theory models to the EnHANTs domain.
6.1 Deterministic Model
Our first example considers a tag which harvests energy from
an on/off periodic energy source shown on the the upper graph of
Figure 4. Such an on/off source can be found, for example, in an
office environment where an indoor lighting system is turned on
in the morning and off at night. The source is on during the period Tp , in which the tag charges at a constant rate r and it is off
during Td . Throughout both periods the tag consumes energy at a
constant rate ec . In the inventory management domain this scenario
directly matches the classic Economic Production Quantity (EPQ)
model [32,33]. It can be easily shown that in this model the highest
consumption rate that can be maintained is ec = Tp · r/(Td + Tp ).
For the EnHANTs domain, the consumption rate can be directly
translated to bit rate, duty cycle, or level of communications.
Consider a scenario where a tag with a 10cm2 solar cell of 1%
efficiency is located on a dimly lit shelf in an enclosed office, where
the irradiance is 50µW/cm2 when the office light is on and 0 when
it is off. If the light in this office is turned on for 10 hours per day,
the tag can spend energy at a constant rate of 50 · 10−6 · 10 · 0.01 ·
10/24 = 2.08µW. Assuming that reception of one bit requires 1nJ
(similar to section 3.1), this tag will be able to maintain the data
rate of 2.08Kb/s throughout the entire diurnal office cycle.
Figure 4 shows an example of all the parameters of this system
which directly correspond to the EPQ model’s dynamics of the supply, the demand, and the inventory in a warehouse as well as in a
tag. The EPQ model is very simple, yet it demonstrates an encouraging resemblance of a tag’s energy harvesting system and a
large-scale factory-warehouse system.
6.2 Stochastic Model
For environments where energy sources are not deterministic,
other models are needed. An example of an inventory model that
E(t)
Battery level
e(t)
r(t)
Energy source
Energy spending
0
2000
4000
6000
Q
s
8000
10000
ec
emin
Figure 5: Application of the order-point, order quantity (s, Q) inventory model to the EnHANTs domain: (i) energy charge rate
r(t), (ii) the energy level E(t), and (iii) the energy consumption
rate e(t).
applies directly to the EnHANTs domain is the order-point, orderquantity (s, Q) model [32, 33], which takes the stochastic nature
of demand for inventory into account. To avoid shortages, in this
model the level of inventory is tracked, and when it falls below a
predetermined level s, additional Q items are ordered. This results
in the inventory level curve similar to the middle graph in Figure 5.
The following EnHANT energy spending policy, similar to the
“battery-state-based” strategy described in [22], results in the same
inventory level (battery level) dynamics. A tag spends energy at a
constant rate ec , but if the tag’s battery level drops below a predetermined value s, the tag switches to a “safety” mode in which it
spends energy at a rate not exceeding a minimal rate emin . The values for s and emin should be selected such that a tag in the “safety”
mode is able to function at some level, for example pair with a few
of its neighbors. The tag stays in the minimum-spending mode until its battery level reaches Q. Then, it returns to spending energy
at its normal ec rate. An example of applying the (s, Q) policy to a
tag with a stochastic source is shown in Figure 5.
6.3 Novel Energy-Inventory Models
While, as shown above, certain inventory management models
directly map to the EnHANTs domain, EnHANTs and, particularly, networks of EnHANTs call for extensions of existing inventory management models. Compared to the inventory management
domain, in the EnHANTs domain the environment is more random
with fewer parameters known with certainty and fewer parameters
under control. For example, while warehouse capacity is usually
known and constant, a tag’s storage capacity C may not be accurately known and may change over time. Also, an EnHANT cannot
manufacture more inventory (energy) when needed, and does not
control the cycles of production. Moreover, in inventory management, the demand is usually stochastic but the supply is mostly deterministic. In the EnHANTs domain, both the supply (energy harvested) and the demand (communications) are likely to be stochastic.3 Existing inventory models need to be extended to take into
account this uncertainty and randomness.
A particular challenge in EnHANTs is the dependency of energy
spending of different communicating tags. A tag should spend energy on transmissions only if the tag it communicates with is ready
to spend energy on reception. Hence, one tag’s energy spending necessitates another tag’s energy spending. Moreover, the transition
of different tags between states are somewhat correlated. Hence,
the resulting system can be viewed as a network of factories in
3 However, as can be seen in Figure 5, the tag can somewhat control
its inventory level by adjusting its energy spending rate.
which the behavior of one significantly affects the behavior of the
others. While some inventory theory models consider multiple
warehouses (multi-echelon models), they will need to be extended
to capture the complexity of EnHANTs. Further, while in a manufacturing system the central controller has complete knowledge, in
a network of tags, distributed low complexity algorithms using partial knowledge will have to be employed. Extending inventory management models to handle the unpredictability of EnHANTs and
EnHANTs dependencies, and creating realistic and implementable
EnHANTs energy spending algorithms is an exciting challenge.
7. TESTBED DESIGN
Experimentation with the various device designs and algorithms
in real-world settings is crucial in order to better understand the design considerations. However, wireless mote designs are optimized
for wireless sensor applications and typically use IEEE 802.15.4
RF transceivers [14, 16]. These transceivers do not leave room
for experimentation with physical layer communications protocols.
Hence, we are in the process of building EnHANTs prototypes.
In the first phase, these prototypes will be based on commercial
off-the-shelf (COTS) components. Currently, they are physically
much larger and consume more power than the targeted EnHANT.
They do not include a UWB transceiver, flexible solar cell, and
a custom battery but will serve as a platform for preliminary experiments. One prototype is based on a MICA2 mote attached
to a TAOS TSL230rd light-to-frequency converter for light measurements and to an Si solar cell for energy harvesting. Another
prototype is based on a LabJack and a TAOS TSL230rd light-tofrequency converter. Using these prototypes, we have been performing energy harvesting measurements in various environments.
These measurements will enable us to develop real-world energy
supply models that will support the development of the algorithms
described in Sections 5 and 6. In addition, we have been using the
MICA2-based prototype to emulate harvesting-aware communications protocols.
In the next phase, we will replace the COTS components with
custom designed hardware (flexible harvester similar to the one in
Figure 2 and a UWB transceiver). This platform will allow us to
demonstrate initial feasibility for the transceiver and harvester and
to test networking protocols with the actual energy budgets. Further
testbed information and results will be available at [1].
8. CONCLUSIONS
We believe that ultra-low-power Energy Harvesting Active Networked Tags (EnHANTs) are enablers for a new type of a wireless
network which lies in the domain between sensor networks and
RFIDs. While RFIDs make it possible to identify an object which
is in proximity to a reader, EnHANTs make it possible to search
for an object on a network of devices and continuously monitor
objects’ locations and proximity to each other. EnHANTs enable
novel tracking applications such as recovery of lost items, locating items with particular characteristics, continuous monitoring of
merchandize, and assistance in locating survivors of a disaster.
EnHANTs necessitate rethinking of communication and networking principles, and require careful examination of the particularities
of ultra-low-power and energy harvesting technologies. We have
shown that the nature of EnHANTs requires a cross-layer approach
to enable effective communications and networking between devices with severe power and harvesting constraints. In order to
discuss the design challenges, we outlined several important characteristics of EnHANTs, pointed out a number of open problems
and possible research directions, and introduced inventory control
theory as a tool that might be used to address open problems related
to the tags’ energy management.
[23]
9.
ACKNOWLEDGMENTS
This work was supported in part by the Vodafone Americas Foundation Wireless Innovation Project, by Google, and by an NSERC
CGS grant. We thank Prof. Steenkiste and the reviewers for their
helpful comments.
[24]
[25]
10. REFERENCES
[1] Energy-Harvesting Active Networked Tags (EnHANTs) Project,
Columbia University, http://enhants.ee.columbia.edu.
[2] GreenPeak Technologies, http://www.greenpeak.com/.
[3] Texas Instruments MSP430 Solar Energy Harvesting Development
Tool, http://focus.ti.com/docs/toolsw/folders/print/
ez430-rf2500-seh.html.
[4] IEEE std 802.15.4a-2007 (amendment to IEEE std 802.15.4-2006).
[5] N. Bambos, “Toward power-sensitive network architectures in
wireless communications: concepts, issues, and design aspects,”
IEEE Pers. Commun., vol. 5, no. 3, pp. 50–59, June 1998.
[6] K.-W. Fan, Z. Zheng, and P. Sinha, “Steady and fair rate allocation
for rechargeable sensors in perpetual sensor networks,” in Proc.
ACM SenSys’08, Nov. 2008.
[7] M. Green, K. Emery, Y. Hishikawa, and W. Warta, “Solar cell
efficiency tables (version 32),” Progress in Photovoltaics: Research
and Applications, vol. 16, no. 5, pp. 435–440, 2008.
[8] J. Hsu, S. Zahedi, A. Kansal, M. Srivastava, and V. Raghunathan,
“Adaptive duty cycling for energy harvesting systems,” in Proc.
IEEE ISLPED’06, Oct. 2006.
[9] N. Jaggi, K. Kar, and A. Krishnamurthy, “Rechargeable sensor
activation under temporally correlated events,” in Proc. IEEE
WiOpt’07, 2007.
[10] C. E. Jones, K. M. Sivalingam, P. Agrawal, and J. C. Chen, “A
survey of energy efficient network protocols for wireless networks,”
ACM/Kluwer Wireless Networks, vol. 7, no. 4, pp. 343–358, 2001.
[11] A. Kansal, J. Hsu, S. Zahedi, and M. B. Srivastava, “Power
management in energy harvesting sensor networks,” ACM Trans.
Embedded Comput. Syst., vol. 6, no. 4, 2007.
[12] A. Kansal, D. Potter, and M. B. Srivastava, “Performance aware
tasking for environmentally powered sensor networks,” in Proc.
ACM SIGMETRICS’04, June 2004.
[13] K. Kar, A. Krishnamurthy, and N. Jaggi, “Dynamic node activation
in networks of rechargeable sensors,” IEEE/ACM Trans. Netw.,
vol. 14, no. 1, pp. 15–26, 2006.
[14] H. Karl and A. Willig, Protocols and Architectures for Wireless
Sensor Networks. Wiley, 2007.
[15] M. Kodialam and T. Nandagopal, “Fast and reliable estimation
schemes in RFID systems,” in Proc. ACM MobiCom’06, Sept. 2006.
[16] B. Krishnamachari, Networking Wireless Sensors. Cambridge
University Press, 2005.
[17] J. Kymissis, C. Kendall, J. Paradiso, and N. Gershenfeld, “Parasitic
power harvesting in shoes,” in Proc. Second Int. Symp. Wearable
Computers, 1998.
[18] L. Lin, N. Shroff, and R. Srikant, “Asymptotically optimal
power-aware routing for multihop wireless networks with renewable
energy sources,” in Proc. IEEE INFOCOM’05, Mar. 2005.
[19] W. Ma, C. Yang, X. Gong, K. Lee, and A. Heeger, “Thermally
stable, efficient polymer solar cells with nanoscale control of the
interpenetrating network morphology,” Advanced Functional
Materials, vol. 15, no. 10, pp. 1617–1622, 2005.
[20] R. Merz, J. Widmer, J.-Y. Le Boudec, and B. Radunovic, “A joint
PHY/MAC architecture for low-radiated power TH-UWB wireless
ad-hoc networks,” Wireless Commun. and Mobile Comp., vol. 5,
no. 5, pp. 567–580, Aug. 2005.
[21] V. Mhatre and C. Rosenberg, “Energy and cost optimizations in
wireless sensor networks: A survey,” in Performance Evaluation and
Planning Methods for the Next Generation Internet. Kluwer, 2005.
[22] D. Niyato, E. Hossain, and A. Fallahi, “Sleep and wakeup strategies
in solar-powered wireless sensor/mesh networks: performance
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
analysis and optimization,” IEEE Trans. Mobile Comput., vol. 6,
no. 2, pp. 221–236, Feb. 2007.
D. Niyato, E. Hossain, M. Rashid, and V. Bhargava, “Wireless
sensor networks with energy harvesting technologies: a
game-theoretic approach to optimal energy management,” IEEE
Wireless Commun., vol. 14, no. 4, pp. 90–96, Aug. 2007.
J. Paradiso and T. Starner, “Energy scavenging for mobile and
wireless electronics,” IEEE Pervasive Comput., vol. 4, no. 1, pp.
18–27, 2005.
P. Peumans, A. Yakimov, and S. Forrest, “Small molecular weight
organic thin-film photodetectors and solar cells,” J. Applied Physics,
vol. 93, p. 3693, 2003.
T. Phan, J. Lee, V. Krizhanovskii, S. Han, and S. Lee, “A 18-pJ/pulse
OOK CMOS transmitter for multiband UWB impulse radio,” IEEE
Microw. Wireless Compon. Lett., vol. 17, no. 9, pp. 688–690, 2007.
M. Rahimi, H. Shah, G. Sukhatme, J. Heideman, and D. Estrin,
“Studying the feasibility of energy harvesting in a mobile sensor
network,” in Proc. IEEE ICRA’03, Sept. 2003.
M. Raju, “Energy harvesting. ULP meets energy harvesting: a
game-changing combination for design engineers,” TI,
http://focus.ti.com/lit/wp/slyy018/slyy018.pdf, 2008.
J. Randall, Designing Indoor Solar Products, 1st ed. Wiley, 2005.
S. Roudy and L. Frechette, “Energy scavenging and nontraditional
power sources for wireless sensor networks,” in Handbook of Sensor
Networks: Algorithms and Architectures, I. Stojmenovic, Ed.
Wiley, 2005.
Y. Shi, Y. T. Hou, H. D. Sherali, and S. F. Midkiff, “Cross-layer
optimization for routing data traffic in UWB-based sensor
networks,” in Proc. ACM MobiCom’05, Sept. 2005.
E. A. Silver, D. F. Pyke, and R. Peterson, Inventory Management and
Production Planning and Scheduling, 3rd ed. Wiley, 1998.
D. Simchi-Levi, X. Chen, and J. Bramel, The Logic Of Logistics:
Theory, Algorithms, and Applications for Logistics and Supply
Chain Management, 2nd ed. Springer, 2005.
M. Tacca, P. Monti, and A. Fumagalli, “Cooperative and reliable
ARQ protocols for energy harvesting wireless sensor nodes,” IEEE
Trans. Wireless Commun., vol. 6, no. 7, pp. 2519–2529, July 2007.
J. Taneja, J. Jeong, and D. Culler, “Design, modeling, and capacity
planning for micro-solar power sensor networks,” in Proc. IEEE
IPSN’08, Apr. 2008.
T. Terada, S. Yoshizumi, M. Muqsith, Y. Sanada, and T. Kuroda, “A
CMOS ultra-wideband impulse radio transceiver for 1-Mb/s data
communications and ś2.5-cm range finding,” IEEE J. Solid-State
Circuits, vol. 41, no. 4, pp. 891–898, Apr. 2006.
C. Vigorito, D. Ganesan, and A. Barto, “Adaptive control of duty
cycling in energy-harvesting wireless sensor networks,” in Proc.
IEEE SECON’07, June 2007.
T. Voigt, A. Dunkels, J. Alonso, H. Ritter, and J. Schiller,
“Solar-aware clustering in wireless sensor networks,” in Proc. IEEE
ISCC’04, 2004.
T. Voigt, H. Ritter, and J. Schiller, “Utilizing solar power in wireless
sensor networks,” in Proc. IEEE LCN’03, Oct. 2003.
R. Want, “An introduction to RFID technology,” IEEE Pervasive
Comput., vol. 5, no. 1, pp. 25–33, 2006.
B. Warneke, M. Last, B. Liebowitz, and K. Pister, “Smart dust:
communicating with a cubic-millimeter computer,” IEEE Computer,
vol. 34, no. 1, pp. 44–51, Jan 2001.
D. Wentzloff, F. Lee, D. Daly, M. Bhardwaj, P. Mercier, and
A. Chandrakasan, “Energy efficient pulsed-UWB CMOS circuits
and systems,” in Proc. IEEE ICUWB’07, Sept. 2007.
M. Z. Win and R. A. Scholtz, “Ultra-wide bandwidth time-hopping
spread-spectrum impulse radio for wireless multiple-access
communications,” IEEE Trans. Commun., vol. 48, no. 4, pp.
679–691, Apr. 2000.
J. Zheng and T. Jow, “High energy and high power density
electrochemical capacitors,” J. Power Sources, vol. 62, no. 2, pp.
155–159, 1996.
G. Zussman and A. Segall, “Energy efficient routing in ad hoc
disaster recovery networks,” Ad Hoc Networks, vol. 1, no. 4, pp.
405–421, Nov. 2003.